Cp - cpk for screwdriving process

But does the machine use the same driver for all 5 screws? Or does it use 5 different drivers? Does it collect the torque data (like an Atlas-Copco system)? Then it would be easy to determine capability for all 5.

 

If you are going to perform a 100-piece capability study, I recommend measuring 20-pieces from each screw. Plot the data on an I-MR control chart and look for any signs of an out-of-control process, particularly shifts in the mean when transitioning from one screw to the next. I would also run an ANOM (Analysis of Means) on the data to see if that shows a difference between screws. If the process is in control and the ANOM does not show a difference, you can combine the results in your capability study AND be confident that there are no differences.

 

It may not make a difference, but then again it may. A simple study such as I described would answer that question definitively. I've seen a lot a strange things in my career that flew in opposition to conventional wisdom.

 

What Miner said.

The problem here as that the screws which are in different positions may be different because of the location - this is called homogeneity. If they are homogenous you can combine the results. (This is what Miner is asking you to test for.). If they not homogenous (different average value from each other) you cannot directly combine them because the difference is ‘fixed’. If you were to combine them then the fixed difference would be seen mathematically as if it were random but it’s not - the formula will give you a invalid result - and more importantly you would be missing a significant source of variation.

Statistics is just as complex and deep as electronic engineering - I know, I have degrees and 40 years of experience in both. I too have seen crazy things: once I had 4 machines that were substantially different from each other although they were the same brand, model and their SNs were sequential. The difference? The location changed the hydraulic pressures just enough because the lines were different lengths.

So, unless you want an in depth statistics lesson, you should follow Miner’s advice. It would be helpful to post your results here so we can help you interpret the results. Statistics isn’t a shake and bake oven…